2026-05-29

ProRL: Effective Reinforcement Learning for Proactive Recommendation via Rectified Policy Gradient Estimation

👍 76

05/27 08:00

Proactive Recommender Systems (PRSs) aim to guide user preference shift toward target items by generating paths of intermediate recommendations. Reinforcement learning (RL) provides a principled framework for optimizing such sequential decision tasks, as path rewards can naturally capture both short

中文介绍 ProRL提出修正策略梯度估计方法，用于主动推荐系统（PRS）中强化学习优化路径生成。通过引导用户偏好向目标物品转移，缓解探索-利用不平衡，提升推荐效果。

Agent Explorative Policy Optimization for Multimodal Agentic Reasoning

👍 71

05/27 08:00

Vision-language models with extended reasoning succeed on complex problems, but many real-world problems require external tools that internal reasoning alone often cannot resolve. Agentic reasoning therefore interleaves two behaviors with a structural asymmetry: thinking (the self-contained default)

中文介绍 Agent Explorative Policy Optimization（AEPO）针对多模态具身推理中思考与工具使用的不对称性，设计探索性策略优化，使VLM更有效地调用外部工具，在复杂任务上提升性能。

Self-Improving Language Models with Bidirectional Evolutionary Search

👍 44

05/27 08:00

Search has been proposed as an effective method for self-improving language models and agentic systems, both for post-training sample generation and for inference. However, widely used methods such as best-of-N sampling and tree search face two fundamental limitations: they are guided by sparse veri

中文介绍 双向进化搜索（Bidirectional Evolutionary Search）用于语言模型自我改进，克服best-of-N和树搜索依赖稀疏验证信号的局限，生成更高质量训练样本和推理路径，提升模型性能。

DenoiseRL: Bootstrapping Reasoning Models to Recover from Noisy Prefixes

👍 39

05/27 08:00

Reinforcement learning has become a central paradigm for advancing reasoning in large language models, yet most existing methods still depend on stronger teacher models or heavily curated difficult datasets, limiting scalable capability improvement. In this paper, we introduce DenoiseRL, a reinforce

中文介绍 DenoiseRL提出强化学习框架，让大语言模型在推理时从噪声前缀中恢复，无需强教师模型或精心筛选的困难数据集。通过去噪训练自举提升推理能力，实现可扩展性能改进。

GEM: Generative Supervision Helps Embodied Intelligence

👍 34

05/27 08:00

Embodied Vision-Language Models (VLMs) have demonstrated impressive performance and generalization in robotics, particularly within Vision-Language-Action frameworks. However, a significant gap remains between the high-level semantic focus of standard text-guided pre-training paradigms and the low-l

中文介绍 GEM提出生成式监督方法，弥合VLA模型中高层语义预训练与低层动作执行的差距。通过生成式任务训练使具身VLM更好理解细粒度物理交互，提升机器人操作泛化能力。

MemTrace: Tracing and Attributing Errors in Large Language Model Memory Systems

👍 33

05/27 08:00

Memory is essential for enabling large language models to support long-horizon reasoning, yet existing memory systems remain unreliable and difficult to debug. Tracing memory's dynamic evolution is crucial to understand how information is synthesized, propagated, or corrupted over time. In this work

中文介绍 MemTrace是用于大语言模型记忆系统的错误追踪与归因工具。通过动态追踪记忆演化，分析信息如何合成、传播或腐化，帮助定位记忆系统可靠性问题，增强长程推理调试能力。

Learn from Weaknesses: Automated Domain Specialization for Small Computer-Use Agents

👍 32

05/27 08:00

Computer-use agents (CUAs) have recently made substantial progress, but deploying a separate large expert for each software domain remains expensive. Small open computer-use agents are more practical specialization targets, but they remain substantially weaker and exhibit uneven domain-specific fail

中文介绍 提出从弱点学习的自动化领域专业化方法，针对小型计算机使用代理（CUA）。自动识别领域特定失败模式并针对性优化，使小模型在目标软件领域显著提升性能，降低部署成本。

ScientistOne: Towards Human-Level Autonomous Research via Chain-of-Evidence

👍 29

05/25 08:00

Autonomous research agents produce competitive solutions and professional-looking manuscripts, yet their outputs contain verifiability failures undetectable by surface-level evaluation: fabricated citations, unreproducible scores, and method descriptions that diverge from the implementation. We addr

中文介绍 ScientistOne提出“证据链”（Chain-of-Evidence）框架，解决自主研究代理输出不可验证问题（虚构引用、分数不可复现等）。通过强制每步结论附证据链，提升研究结果可信度和可复现性。

AI Research Agents Narrow Scientific Exploration

👍 20

05/27 08:00

AI research agents can now generate research ideas, design experiments, run code, and draft papers, raising the possibility of large-scale AI-assisted scientific discovery. Many current agent frameworks explicitly encourage the generation of novel and high-impact ideas. Yet it remains unclear whethe

中文介绍 研究发现AI研究代理会收窄科学探索范围。尽管代理能生成新颖想法，但倾向聚焦已知流行方向，减少对冷门但有潜力领域的探索，警示大规模使用AI辅助科研时的探索偏差。

Rethinking Memory as Continuously Evolving Connectivity

👍 19

05/27 08:00

Existing memory-augmented LLM agents often treat memory as a static repository with pre-defined representations and fixed retrieval pipelines, which is brittle in dynamic agentic environments where feedback, task variation, and heterogeneous signals continuously reshape what should be remembered and

中文介绍 提出将记忆视为持续演化的连接性网络，替代传统静态记忆存储。通过动态更新记忆结构以适应环境反馈和任务变化，增强LLM代理在动态场景下的适应性和鲁棒性。

Triplet-Block Diffusion RWKV

👍 18

05/25 08:00

Causal Transformer language models suffer from strictly sequential decoding and a quadratic per-step attention cost. While linear-time causal models and discrete diffusion models each address these weaknesses, their integration remains inherently inconsistent: diffusion requires bidirectional attent

中文介绍 Triplet-Block Diffusion RWKV融合线性时间因果模型与离散扩散模型，解决Transformer推理慢和二次注意力成本问题。通过三块架构设计实现双向扩散与因果生成的一致整合，提升生成效率。

GUI-CIDER: Mid-training GUI Agents via Causal Internalization and Density-aware Exemplar Reselection

👍 16

05/27 08:00

Despite the rapid progress of multimodal large language models in building Graphical User Interface (GUI) agents, their real-world task completion is fundamentally bottlenecked by a lack of world knowledge about GUI operations. Existing solutions typically rely on expensive multi-agent scaffolding o

中文介绍 GUI-CIDER通过因果内化（Causal Internalization）和密度感知样本重选（Density-aware Exemplar Reselection）进行中期训练，注入GUI操作世界知识，无需昂贵多智能体框架，显著提升GUI代理在真实任务中完成率。

Long Live The Balance: Information Bottleneck Driven Tree-based Policy Optimization

👍 16

05/27 08:00

Recent advances in online reinforcement learning (RL) for large language models (LLMs) have demonstrated promising performance in complex reasoning tasks. However, they often exhibit an imbalanced exploration-exploitation trade-off, resulting in unstable optimization and sub-optimal performance. We

中文介绍 提出信息瓶颈驱动的树搜索策略优化方法，用于LLM在线强化学习。通过信息瓶颈准则平衡探索与利用，减少不稳定优化，提升复杂推理任务性能。

SkillGrad: Optimizing Agent Skills Like Gradient Descent

👍 15

05/26 08:00

Agent skills provide a lightweight way to adapt LLM agents to specialized domains by storing reusable procedural knowledge in structured files. However, whether downloaded from third parties or self-generated, these skills are often unreliable, incomplete, or outdated. Existing skill-evolution metho

中文介绍 SkillGrad将智能体技能优化类比为梯度下降，通过迭代更新技能文件中的程序知识。解决第三方或自生成技能不可靠、不完整或过时问题，使LLM代理更高效适应专业领域。

Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving

👍 15

05/25 08:00

End-to-end autonomous driving via Vision-Language-Action (VLA) models demands a precarious balance between high-fidelity trajectory planning and efficient inference. Existing paradigms typically fall short: autoregressive (AR) VLAs are memory-bandwidth-bound on edge hardware and prone to exposure-bi

中文介绍 Fast-dDrive提出块扩散VLA模型，在自动驾驶中实现高保真轨迹规划与高效推理的平衡。通过块扩散避免自回归模型的带宽瓶颈和曝光偏差，加速边缘部署。

Your Agents Are Aging Too: Agent Lifespan Engineering for Deployed Systems

👍 14

05/25 08:00

Long-lived AI agents are increasingly deployed as persistent operational systems, yet they are still evaluated like freshly initialized models. Day-one benchmarks miss a basic systems question: how long does an agent remain reliable after deployment? Even when model weights are frozen, an agent's ef

中文介绍 提出智能体寿命工程（Agent Lifespan Engineering），研究部署后代理可靠性衰减问题。即使模型权重固定，代理效率也因环境变化而下降，通过生命周期评估与维护策略延长有效运行时间。

GE-Sim 2.0: A Roadmap Towards Comprehensive Closed-loop Video World Simulators for Robotic Manipulation

👍 13

05/26 08:00

We introduce GE-Sim 2.0 (Genie Envisioner World Simulator 2.0), a closed-loop video world simulator for robotic manipulation. Building on the action-conditioned video generation framework of Genie Envisioner, GE-Sim 2.0 is re-trained on thousands of hours of real-world robot data spanning teleoperat

中文介绍 GE-Sim 2.0是一个闭环视频世界模拟器，基于动作条件视频生成框架，利用数千小时真实机器人数据重新训练。支持遥操作数据，为机器人操作提供高保真交互式仿真环境。

Guiding LLM Post-training Data Engineering with Model Internals from Sparse Autoencoders

👍 12

05/26 08:00

Model internals encode rich information about how a large language model (LLM) processes its training data; however, post-training data engineering largely relies on external signals and ignores rich intrinsic signals lying in model internals. We propose SAERL, a data engineering framework for LLM r

中文介绍 SAERL利用稀疏自编码器提取模型内部表征，指导LLM后训练数据工程。通过分析模型如何处理数据，优化数据选择与混合策略，提升后训练效果，减少对外部信号的依赖。

LiveBrowseComp: Are Search Agents Searching, or Just Verifying What They Already Know?

👍 11

05/27 08:00

Are LLM-based search agents genuinely searching, or using the web to verify what they already know? We study this question on BrowseComp with three diagnostics. Our analysis reveals Intrinsic Knowledge Dependence (IKD): even with tool access, agents often rely on intrinsic knowledge -- information e

中文介绍 LiveBrowseComp通过三种诊断方法分析搜索代理是否存在内在知识依赖（IKD）：即使能使用工具，代理仍倾向依靠自身已有知识而非真正搜索，揭示验证性行为，推动更可靠搜索设计。

HRBench: Benchmarking and Understanding Thinking-Mode Switch Strategies in Hybrid-Reasoning LLMs

👍 10

05/27 08:00

Hybrid-reasoning large language models (LLMs) expose explicit controls over reasoning effort, allowing users or systems to trade off answer quality against inference cost. However, existing methods for adaptive thinking-mode selection are typically evaluated under different models, datasets, and imp

中文介绍 HRBench提供混合推理LLM的思考模式切换策略基准。统一评估不同模型和数据集上快速/深度模式切换效果，量化质量与成本权衡，为自适应推理提供标准化测试平台。

The Age of Async Agents — Cognition's Walden Yan & OpenInspect's Cole Murray

Latent Space · 05/29 02:41

80% Devin Commits, Spec-to-PR Workflows, Full VMs, Agent Memory, and PMs Shipping Code

中文介绍 Cognition公司的Walden Yan与OpenInspect的Cole Murray讨论异步代理时代，内容包括Devin提交占比80%、规约到PR工作流、完整虚拟机、代理内存以及产品经理发布代码。

The AI Hype Index: AI gets booed in graduation season

MIT Tech Review AI · 05/28 17:51

It is one thing to say AI will change the world. It is another to expect the class of 2026 to applaud it. In fact, when former Google CEO Eric Schmidt told University of Arizona graduates that their task is to help shape AI, he was met with a resounding chorus of boos. “I can…

中文介绍 MIT Tech Review的AI炒作指数显示，毕业季AI遭到嘘声。前谷歌CEO Eric Schmidt在亚利桑那大学演讲时告诉毕业生要帮助塑造AI，却遭遇一片嘘声。2026届毕业生对AI态度冷淡。

[AINews] Cognition raises $1B in $26B Series D

Latent Space · 05/28 15:26

coding is an uncapped TAM market

中文介绍 Cognition完成D轮融资，筹集10亿美元，估值达260亿美元。报道指出编程市场是无限的总可寻址市场（TAM）。

OpenAI’s Frontier Governance Framework

OpenAI News · 05/28 08:00

Explore OpenAI’s Frontier Governance Framework and how our AI safety, security, and risk practices align with emerging EU and California regulations.

中文介绍 OpenAI发布前沿治理框架，阐述其AI安全、安保和风险实践如何与欧盟和加州新兴法规保持一致。

OpenAI private MCP 🤖, Cognition $26B valuation 💰, ElevenLabs Music v2 🎵

TLDR AI · 05/28 08:00

中文介绍 OpenAI推出私有MCP；Cognition估值达260亿美元；ElevenLabs发布Music v2。

🔬ESM: The Bitter Lesson is Coming for Proteins - Alex Rives, BioHub

Latent Space · 05/28 01:46

Biohub’s Protein World Model: ESMC-6B, ESMFold2, 6.8B proteins, 1.1B structures, antibody design, SAEs, & the potential for programmable biology

中文介绍 Biohub的蛋白质世界模型：ESMC-6B、ESMFold2，涵盖68亿蛋白质、11亿结构、抗体设计、稀疏自动编码器（SAEs）及可编程生物学的潜力。Alex Rives参与讨论。

ITBench-AA: Frontier Models Score Below 50% on the First Benchmark for Agentic Enterprise IT Tasks — by Artificial Analysis and IBM

Hugging Face Blog · 05/28 01:20

中文介绍 Artificial Analysis与IBM联合推出ITBench-AA基准，用于评估代理型企业IT任务。前沿模型得分低于50%。

Cisco and OpenAI redefine enterprise engineering with Codex

OpenAI News · 05/27 19:00

Cisco and OpenAI are redefining enterprise engineering with Codex, helping Cisco scale AI-native development, accelerate AI Defense work, and automate defect remediation.

中文介绍 思科与OpenAI合作，利用Codex重新定义企业工程，帮助思科扩展AI原生开发、加速AI防御工作并自动化缺陷修复。

Building self-improving tax agents with Codex

OpenAI News · 05/27 15:00

See how OpenAI, Thrive, and Crete built a self-improving tax agent with Codex, automating filings, improving accuracy, and accelerating workflows.

中文介绍 OpenAI、Thrive和Crete利用Codex构建自我改进的税务代理，实现申报自动化、提高准确性并加速工作流程。

[AINews] New AI Infra decacorns: Fireworks, Baseten (with OpenRouter on the way)

Latent Space · 05/27 11:33

it's funding news, but it's good news.

中文介绍 AI基础设施领域出现新十角兽：Fireworks和Baseten获得高估值，OpenRouter也在路上。

Election information and safeguards in 2026

OpenAI News · 05/27 08:00

Ahead of global elections, we’re helping people access information, supporting cyber defenders, and increasing AI transparency

中文介绍 OpenAI针对2026年全球选举推出保障措施，帮助人们获取信息、支持网络防御者并提高AI透明度。

Warp’s big bet on building open source with GPT-5.5

OpenAI News · 05/27 08:00

Warp uses GPT-5.5 and OpenAI models to coordinate coding agents across local, cloud, and open-source development workflows.

中文介绍 Warp公司押注开源，利用GPT-5.5和OpenAI模型协调跨本地、云端和开源开发工作流的编程代理。

Reachy Mini goes fully local

Hugging Face Blog · 05/27 08:00

中文介绍 Reachy Mini机器人实现完全本地化运行，无需云端连接。

Shipping a Trillion Parameters With a Hub Bucket: Delta Weight Sync in TRL

Hugging Face Blog · 05/27 08:00

中文介绍 Hugging Face博客介绍TRL中的Delta权重同步技术，可通过Hub Bucket传输万亿参数模型。

xAI Cursor limits 🚫, DeepSWE 👨‍💻, China AI travel restrictions 🤖

TLDR AI · 05/27 08:00

中文介绍 xAI对Cursor设置限制；DeepSWE项目推出；中国实施AI旅行限制。